Matches in SemOpenAlex for { <https://semopenalex.org/work/W3171116780> ?p ?o ?g. }
Showing items 1 to 92 of
92
with 100 items per page.
- W3171116780 abstract "Certifiable local robustness, which rigorously precludes small-norm adversarial examples, has received significant attention as a means of addressing security concerns in deep learning. However, for some classification problems, local robustness is not a natural objective, even in the presence of adversaries; for example, if an image contains two classes of subjects, the correct label for the image may be considered arbitrary between the two, and thus enforcing strict separation between them is unnecessary. In this work, we introduce two relaxed safety properties for classifiers that address this observation: (1) relaxed top-k robustness, which serves as the analogue of top-k accuracy; and (2) affinity robustness, which specifies which sets of labels must be separated by a robustness margin, and which can be $epsilon$-close in $ell_p$ space. We show how to construct models that can be efficiently certified against each relaxed robustness property, and trained with very little overhead relative to gradient descent. Finally, we demonstrate experimentally that these relaxed variants of robustness are well-suited to several significant classification problems, leading to lower rejection rates and higher certified accuracies than can be obtained when certifying standard local robustness." @default.
- W3171116780 created "2021-06-22" @default.
- W3171116780 creator A5057424614 @default.
- W3171116780 creator A5077941255 @default.
- W3171116780 date "2021-06-11" @default.
- W3171116780 modified "2023-09-27" @default.
- W3171116780 title "Relaxing Local Robustness" @default.
- W3171116780 cites W2180612164 @default.
- W3171116780 cites W2460937040 @default.
- W3171116780 cites W2535873859 @default.
- W3171116780 cites W2565186948 @default.
- W3171116780 cites W2594877703 @default.
- W3171116780 cites W2741933435 @default.
- W3171116780 cites W2773726006 @default.
- W3171116780 cites W2786163515 @default.
- W3171116780 cites W2804566015 @default.
- W3171116780 cites W2807040120 @default.
- W3171116780 cites W2892430965 @default.
- W3171116780 cites W2896266676 @default.
- W3171116780 cites W2901458284 @default.
- W3171116780 cites W2911634294 @default.
- W3171116780 cites W2924630351 @default.
- W3171116780 cites W2957311447 @default.
- W3171116780 cites W2962851953 @default.
- W3171116780 cites W2962943487 @default.
- W3171116780 cites W2963207607 @default.
- W3171116780 cites W2963564348 @default.
- W3171116780 cites W2963770044 @default.
- W3171116780 cites W2963952467 @default.
- W3171116780 cites W2964153729 @default.
- W3171116780 cites W2970567221 @default.
- W3171116780 cites W2995053405 @default.
- W3171116780 cites W2996261834 @default.
- W3171116780 cites W3082511295 @default.
- W3171116780 cites W3101965459 @default.
- W3171116780 cites W3118608800 @default.
- W3171116780 cites W3120829775 @default.
- W3171116780 cites W3172170142 @default.
- W3171116780 cites W2551176409 @default.
- W3171116780 hasPublicationYear "2021" @default.
- W3171116780 type Work @default.
- W3171116780 sameAs 3171116780 @default.
- W3171116780 citedByCount "1" @default.
- W3171116780 countsByYear W31711167802021 @default.
- W3171116780 crossrefType "posted-content" @default.
- W3171116780 hasAuthorship W3171116780A5057424614 @default.
- W3171116780 hasAuthorship W3171116780A5077941255 @default.
- W3171116780 hasConcept C104317684 @default.
- W3171116780 hasConcept C11413529 @default.
- W3171116780 hasConcept C119857082 @default.
- W3171116780 hasConcept C154945302 @default.
- W3171116780 hasConcept C185592680 @default.
- W3171116780 hasConcept C33923547 @default.
- W3171116780 hasConcept C41008148 @default.
- W3171116780 hasConcept C55493867 @default.
- W3171116780 hasConcept C63479239 @default.
- W3171116780 hasConceptScore W3171116780C104317684 @default.
- W3171116780 hasConceptScore W3171116780C11413529 @default.
- W3171116780 hasConceptScore W3171116780C119857082 @default.
- W3171116780 hasConceptScore W3171116780C154945302 @default.
- W3171116780 hasConceptScore W3171116780C185592680 @default.
- W3171116780 hasConceptScore W3171116780C33923547 @default.
- W3171116780 hasConceptScore W3171116780C41008148 @default.
- W3171116780 hasConceptScore W3171116780C55493867 @default.
- W3171116780 hasConceptScore W3171116780C63479239 @default.
- W3171116780 hasLocation W31711167801 @default.
- W3171116780 hasOpenAccess W3171116780 @default.
- W3171116780 hasPrimaryLocation W31711167801 @default.
- W3171116780 hasRelatedWork W2182127273 @default.
- W3171116780 hasRelatedWork W2768915615 @default.
- W3171116780 hasRelatedWork W2913881544 @default.
- W3171116780 hasRelatedWork W2943368551 @default.
- W3171116780 hasRelatedWork W2945557596 @default.
- W3171116780 hasRelatedWork W2947251152 @default.
- W3171116780 hasRelatedWork W2949277570 @default.
- W3171116780 hasRelatedWork W2979946977 @default.
- W3171116780 hasRelatedWork W3106345340 @default.
- W3171116780 hasRelatedWork W3127699700 @default.
- W3171116780 hasRelatedWork W3130306060 @default.
- W3171116780 hasRelatedWork W3139282803 @default.
- W3171116780 hasRelatedWork W3139639917 @default.
- W3171116780 hasRelatedWork W3160187677 @default.
- W3171116780 hasRelatedWork W3175418069 @default.
- W3171116780 hasRelatedWork W3203257194 @default.
- W3171116780 hasRelatedWork W3213376958 @default.
- W3171116780 hasRelatedWork W3214397047 @default.
- W3171116780 hasRelatedWork W753876003 @default.
- W3171116780 hasRelatedWork W2181917792 @default.
- W3171116780 isParatext "false" @default.
- W3171116780 isRetracted "false" @default.
- W3171116780 magId "3171116780" @default.
- W3171116780 workType "article" @default.